Antonio González-Hernández · Rene Morales-Villafa?a ·Martin Enrique Romero-Sánchez · Brenda Islas-Trejo · Ramiro Pérez-Miranda

Abstract Accurate and reliable predictions of pest species distributions in forest ecosystems are urgently needed by forest managers to develop management plans and monitor new areas of potential establishment. Presence-only species distribution models are commonly used in these evaluations. The maximum entropy algorithm (MaxEnt) has gained popularity for modelling species distribution. Here,MaxEnt was used to model the spatial distribution of the Mexican pine bark beetle (Dendroctonus mexicanus) in a daily fashion by using forecast data from the Weather Research and Forecasting model. This study aimed to exploit freely available geographic and environmental data and software and thus provide a pathway to overcome the lack of costly data and technical guidance that are a challenge to implementing national monitoring and management strategies in developing countries. Our results showed overall agreement values between 60 and 87%. The results of this research can be used for D. mexicanus monitoring and management and may aid as a model to monitor similar species.

Keywords Spatial analysis · Dendroctonus mexicanus ·Geodatabases · MaxEnt · Forest modelling

Introduction

Forest conservation has become a priority in the last decade because of the importance of forests in mitigating climate change (Dale et al. 2001). Forest ecosystems provide multiple and critical ecosystem services that are essential to sustaining human well-being (Matthews et al. 2013).However, environmental alterations caused mainly by human activities and to recent events associated with climate change (Sáenz-Romero et al. 2012) may lead to disturbances that compromise the structure, functionality and spatial distribution of forests. A variety of disturbances,either natural or human-induced decrease the productivity and extent of forests (Baker 1995; Dale et al. 2001).

Bark beetle infestations are a leading cause of disturbances, especially in temperate forests, and affect commercial timber production (Salda?a 1989). Climatic anomalies produced by climate change have the most significant impact on outbreaks of bark beetles, and Mexico is one of the countries that are predicted to be more affected by climate change (IPCC 2014). As a consequence, insect outbreaks and wildfires will be more frequent, especially in temperate forests (Matthews et al. 2013).

Although insects are essential in the forest as phytophages, decomposers, pollinators, predators, and parasites, it is imperative to keep the populations under control.Insects in the order Coleoptera (beetles) stand out as the most important due to their number of species and their impact on natural forest dynamics (Evangelista et al. 2011).Beetles feed on decaying wood, and others are predators and control populations of other species. Bark beetles attack trees when conditions are favourable, like those that after a forest fire or severe drought (Delgado and Pedraza Perez 2002).

Beetles from the genus Dendroctonus usually are considered as among the most aggressive insects because they prefer living trees as hosts to ensure successful reproduction (Coulson and Witter 1984). Bark beetles (Dendroctonus mexicanus Hopkins, 1909) represent an essential factor in the degradation and loss of forest ecosystems in Mexico, with environmental, economic and social implications (Moser et al. 2005; Manzo-Delgado et al. 2014).For this reason, information is needed to develop strategies to guard against possible outbreaks (Salinas-Moreno et al.2004, 2010). For instance, we need to determine the distribution of the target species to identify which forest areas are at risk of infestation, and which are prone to be affected. The presence of bark beetles can be prevented through vigilance since timely detection and proper diagnosis increases the possibility of eradicating and managing the problem species and preventing insect dispersal to other sites (Sánchez-Martínez and Wagner 2009).

Spatial modelling and niche ecological models combined with other analytical tools (e.g., geographic information system [GIS]) offer an option for evaluating noninventoried sites and modelling past and future scenarios of the distribution of bark beetles (Manzo-Delgado et al.2014). Recently, numerous modelling methods and tools have been developed (Guisan and Zimmermann 2000; Ray et al. 2017), but they have mainly been used in ecology and biogeography (Anderson et al. 2003; Phillips et al. 2006).Therefore, the use of these alternatives accomplishes a double function: first, they provide knowledge about the potential distribution of the species to determine the richness and diversity of non-evaluated areas. Second, they can be used to predict choice sites for biological conservation zones (Meggs et al. 2004; Chefaoiu et al. 2005).

In this study, we used a predictive model (MaxEnt algorithm), existing records of the presence of the Mexican bark beetle and environmental information layers to generate bioclimatic profiles and to determine the most likely geographic distribution of the Mexican bark beetle. We estimated the probability of occurrence of the species looking for the most uniform distribution possible (Phillips et al. 2006), in such a way that the model expresses an adequate value of the potential habitat for the species as a function of the environmental variables used. Later, we used forecast values to modify the environmental conditions and fit the climatic conditions through the year. This approach would enable us to predict how the most likely geographic distribution of the Mexican bark beetle(Dendroctonus mexicanus) would change according to climatic responses. Finally, we identified forest areas that have the most suitable conditions for bark beetles.

Materials and methods

Study area

The study area comprised temperate coniferous forest across Mexico (Fig. 1). Mexican temperate forests represent around 13% of the Mexican territory (SEMARNAT 2010) are composed mainly of the genus Pinus (Rzedowski 2006; Colditz et al. 2010) but also include fir, cedar and junipers and are characterised by their particular climate and topographic conditions (Gebhardt et al. 2014). In Mexico, pines are found in three types of communities:pure pine forests, pine-oak and oak-pine forest that occur at different elevations, climates and exposures (Rzedowski 2006), representative of the high ecological diversity of the Mexican landscape. Mexican temperate forests are located mainly in high altitudes over mountainous zones originated by volcanos (Gebhardt et al. 2014). The most typical mountain ranges in Mexico are the Trans-Mexican Volcanic Belt located in the central part of the country: the western Sierra Madre Occidental and eastern Sierra Madre Oriental.

Data

Geographic layers

The National Institute of Statistics and Geography of Mexico (INEGI) produces and publishes land cover and vegetation maps for Mexico at a scale of 1:250,000 using a 25-ha minimum mapping unit (INEGI 2008). From this land cover and vegetation map, the ‘‘temperate coniferous forest’’ land cover class was isolated to delineate the study area and provide a spatial context for the analysis. A national Digital Elevation Model (DEM) raster data set with approximately 30-m pixel size was used to create slope and aspects layers for further analysis. Soils and topography layers at a scale of 1:250,000 were also used to extract information that was incorporated later in the study.All digital geographic data described before are freely available (www.inegi.gob.mx) and were provided by INEGI.

Fig. 1 Temperate coniferous forests in Mexico (INEGI 2015)

Climatic data layers such as mean temperature and precipitation (1902-2011) were retrieved from the Digital Climatic Atlas of Mexico (DCAM), developed by the Informatics Unit for Atmospheric and Environmental Sciences (UNIATMOS) of the National Autonomous University of Mexico (UNAM). We used these data as primary input for modelling spatial distributions of pine bark beetle according to its bioclimatic requirements(Cuéllar et al. 2012). Additional bioclimatic variables were collected from WorldClim (Parra 2015) and GIS. World-Clim is an open climate database of global climate layers(gridded climate data) with a spatial resolution of about 1 km2available at www.worldclim.org.

Pine bark beetle data sets

The National Forestry Commission of Mexico (CONAFOR) provided databases that include geographic records of different bark beetle observations across the country.Field observations from 2011 to 2015 included descriptions of the ecological and environmental conditions for each bark beetle observed. We used the Mexican bark beetle as the target species because of its importance and presence in Mexican forests. It attacks 21 Mexican Pinus species(Salinas-Moreno et al. 2010); the most important are P.leiophylla, P. pseudostrobus, P. teocote, P. devoniana, P.montezumae, P. ocarpa, and P. patula (Table 1). From CONAFOR databases, we selected sites where Dendroctonus mexicanus was identified and geographic coordinates specified. All observations of Mexican bark beetle were incorporated and displayed in a GIS environment for visual analysis. After screening and quality control, geographic locations of Mexican pine bark beetle were transformed to an ESRI (Redlands, CA, USA) shapefile format in ArcGIS?version 10.6 to derive layers for each year of observation.

To create a geodatabase, we used 1710 observations that served as primary input for this study. Among these observations, the model included 1510 of the original observations, and 200 were retained to validate the model.

Modelling current spatial distribution

For predicting the potential distribution of the pine bark beetle, we needed records of beetle presence in decimal degrees (compatible with MaxEnt). At the Geomaticslaboratory of the National Institute of Forestry, Agriculture and Livestock Research (INIFAP) in a preliminary analysis, we transformed coordinates to decimal degrees and obtained shapefiles of real distribution by removing incorrect or corrupted data. Probability maps for the geographic distribution of Dendroctonus mexicanus were generated using the MaxEnt model (Paquit et al. 2017),according to their environmental requirements (Table 2).The MaxEnt software was acquired through the website www.cs.princeton.edu/～schapire/MaxEn; version 3.3.3k is freely accessible for download on the website as a zip file.

Table 1 Incidence (%) of Dendroctonus mexicanus on Pinus host species (modified from Salinas-Moreno et al. 2010)

Table 2 Bioclimatic variables from WorldClim (Hijmans et al.2005)

Modelling spatial distribution according to forecast data

We used the Weather Research and Forecasting (WRF)model (Knievel et al. 2007) to obtain forecast variables through a mobile application developed by the National Laboratory of Modelling and Remote Sensing of INIFAP.The forecasts enabled the second set of predictions on the distribution of D. mexicanus. Based on environmental variables for the first days of each month, we used the algorithm MaxEnt and the WRF model outputs to predict potential areas suitable for outbreaks of the bark beetle.September 2016 was selected to test the approach because the most bark beetle outbreaks have been documented in September (Castellanos Bola?os et al. 2013). Numerical values obtained from the WRF mobile application described earlier were interpolated to create a raster image of the Mexican Republic for each of the variables obtained.Subsequently, data files were converted to ASCII format,required by MaxEnt modelling, to generate the potential distribution of Mexican bark beetle according to forecast values (Phillips et al. 2006). The layers obtained were later compared against the first model and field observations to highlight forest areas that are most likely for possible D.mexicanus outbreaks if the environmental conditions change.

Model validation

Once the layers were created, the area under the curve(AUC) of the receiver operating characteristic plot was used to measure the discrimination capacity for all data sets. The AUC is an independent threshold measure of accuracy that compares the rate of true and false positives of validation data across all available habitat suitability thresholds. Values close to 1 indicate perfect discrimination capacity, values close to 0 indicate poor discrimination capacity, and values close to 0.5 indicate discrimination capacity no higher than random (Ray et al.2017). The AUC calculation was built into the MaxEnt software and was thus performed as a procedural component of the model (Phillips et al. 2006; Paquit et al.2017).

Also, we used the 200 observations registered from January to June 2016 to validate the distribution model independently by comparing the number of observations that matched the potential spatial distribution estimated by MaxEnt. We used a threshold of 80% (or higher) of the probability of occurrence to evaluate the performance of the model compared to the observations in the field.

Results

The potential distribution models generated by MaxEnt for Mexican bark beetles, varied concerning the scope of the area of prediction; however, the model generated AUC values close to 1, which is supported by a fastness statistic.AUC values of 0.5 and close to 1 indicate that the generated model is better than the prediction (Figs. 2, 3).

In this work, the AUC value was taken as a measure of model performance (Phillips et al. 2006). Table 3 summarises the results of applying the MaxEnt model to estimate the potential spatial distribution of D. mexicanus.

Spatial distribution of Dendroctonus mexicanus

The potential spatial distribution generated for the bark beetle is shown in Figs. 2 and 3; the higher probabilities of the presence of Mexican bark beetle are in a central area of Mexico constituting the Trans-Mexican Volcanic Belt,where host species such as P. leiophylla, P. pseuodstrobus,P. teocote, among others, are well distributed. Also, the Sierra Madre Occidental, a significant mountain range in western Mexico had areas that qualify as potential areas for D. mexicanus.

The total potential distribution generated with MaxEnt for the Mexican pine bark beetle indicates reliable modelling since the AUC is very high (0.93). The potential distribution for D. mexicanus was based on the relative contributions of the variables used in the modelling.MaxEnt assigns the increase in gain to the environmental variables that are occupied, converts values to percentages and creates a test to estimate the most critical variables in the model (jackknife) (Fig. 4).

Fig. 2 Area under the curve (AUC) plot

Fig. 3 Potential spatial distribution of Dendroctonus mexicanus in Mexican temperate forests

Table 3 Values used for modelling potential distribution

According to results, temperature was the most crucial factor in determining suitable areas for the species. Annual temperature oscillation, mean temperature in the rainiest quarter and in the warmest quarter were among the top contributors as predictor variables (Table 4). However, the variable with the highest gain, when used in isolation, was‘‘bio_10’’ (mean temperature warmest quarter) (Fig. 4),which was considered as the most-informative variable.The variable that decreased the gain in the prediction model was bio_19 (mean precipitation coldest quarter).

It is important to highlight that the five predictors variables comprised 85% of the contributions to the Max-Ent model (Table 4), and four variables related to temperature comprised about 70% of the contributions.

Spatial distribution of Dendroctonus mexicanus based on forecast variables

The Weather Research and Forecasting (WRF) model provided daily values of temperature (maximum, mean and minimum), relative humidity and precipitation for the selected month. The forecasts enabled the second set of predictions about the distribution of D. mexicanus using the algorithm MaxEnt. The results for September 2016 also showed a good fit since the AUC values were high for the days selected (Table 5).

Modelling the spatial distribution of D. mexicanus using forecast data showed that temperature is a critical factor that stimulates the optimal conditions for the development of bark beetles. Again, the significant contributions to the predictive model were made by variables associated with temperature (Table 6).

Fig. 4 Contributions of environmental variables on spatial distribution modelling for the Mexican bark beetle

Table 4 Top predictors and their percentage contribution reported from the Maxent model for Mexican bark beetle

Table 5 AUC values for 6 days of September

Note that we used daily values for precipitation and temperature generated by the forecast model to derive potential distributions, and these predictions could be affected because data were not available for all the environmental variables.

According to results, the potential spatial distribution for September 12-17, 2016 indicated a spatial and temporal relation within the major mountain systems. For instance,the Trans-Mexican Volcanic Belt had the most extensive area of potential distribution for all days. To provide spatial and temporal context to the results, we isolated the Mexican states that comprise the Trans-Mexican Volcanic Belt and calculated the total area to later discuss the changes according to the prediction for each day (Fig. 5).

Figure 5 shows that Mexico State and Michoacan are consistent with the analysis in terms of areas over 500 km2.Also, Mexico State and Michoacán are in the central part of the Trans-Mexican Volcanic Belt where all species of Pinus that serve as hosts of D. mexicanus are naturally distributed (see Table 1).

Table 6 Predictors and their percentage contribution reported from MaxEnt model for Mexican bark beetle using forecast data for September

Fig. 5 Suitable area (km2) for the potential distribution of D. mexicanus for 6 days in September 2016 according to forecast data. The y-axis represents the states within the Trans-Mexican Volcanic Belt

Validation

To compare the observed distribution with the potential distribution generated for each of the 6 days in September,we isolated specific areas where probability was high(＞80%) for potential distributions based on the forecast data. Figure 6 shows the visual comparison of isolated areas for September 2016 against field observations obtained from January to June 2016. It is worth noting the high agreement between field observations and the forecast distributions, thus validating the results obtained (Fig. 6).The potential distributions estimated for the days used in September 2016 compared against the observations collected on the field showed relative agreement for all cases(Table 7).

Discussion

High-density populations of bark beetles have been responsible for economic, environmental and social losses within forest ecosystems (Moser et al. 2005; Manzo-Delgado et al. 2014). Bark beetles are part of the forest dynamic, but lately, as the climate changes, especially with anomalies in temperature (generally high temperatures),conditions are more favourable for the development of bark beetles as a pest of Mexican temperate coniferous forests(IPCC 2014). Of the species reported as a host of D.mexicanus, none is registered at any category for protection(Gernandt and Pérez-de la Rosa 2014); however, all Pinus species are essential for the environmental, economic and social benefits they provide to local communities.

Fig. 6 Visual agreement between potential outbreaks of bark beetle for September 2016 according to climatic forecast data and field observations

Table 7 Cross-tabulation agreement (average accuracy):observed versus estimated

According to the MaxEnt algorithm, Dendroctonus mexicanus showed preferential distribution for temperate,humid and subhumid environments, with an average annual temperature between 12 and 18 °C. Concerning precipitation, the species can be found in areas with 600-1200 mm;however, the potential distribution suggests that the optimal range is 800 mm. During the modelling, we identified temperature and associated variables as crucial in determining the distribution of the Mexican bark beetle,probably because the rate of development of some species of bark beetles in the genus Dendroctonus, including D.mexicanus, is primarily dependent on temperature (Mitton and Sturgeon 1982).

In this study, the primary potential distribution of the target species was located within the temperate coniferous forest in the central part of Mexico, specifically in the Trans-Mexican Volcanic Belt, a portion to the north of the province of the Balsas Depression and part of the Plains and Mountains of Queretaro and Hidalgo. Our results agree with previous reports on the distribution of D. mexicanus in Mexican temperate pine forests in the central region of Mexico (Moser et al. 2005; Sánchez-Martínez and Wagner 2009; Salinas-Moreno et al. 2010; Smith et al. 2013). It is important to highlight that the spatial distribution and number of observations from the field were an essential factor in the performance of the model and the ability to predict the potential spatial distribution of D. mexicanus accurately. Although not part of this study, it would be worthy to evaluate the performance by region according to different number of field observations and spatial configurations.

The approach followed here took advantage of the freely available data to monitoring a critical species such as D.mexicanus and species distribution models (i.e., MaxEnt algorithm). MaxEnt model fits with limited presence-only data provided robust estimates of habitat suitability for species on the landscape as it is confirmed in earlier studies(West et al. 2016). As far as we know, there is no maximum number for the model; however, species distribution models (SDMs) have proved useful performance using a minimum of 14 observations for narrow-ranged species and to 25 for widespread (van Proosdij et al. 2016). In this study, we used 1510 observations from the field, and we assumed that the high number of observations had a positive impact (AUC = 0.93) on the performance of the MaxEnt algorithm.

Although the algorithm does not consider disturbances or habitat fragmentation, areas that are more likely to be suitable for the development of the beetle were those that are suffering from degradation (e.g. disturbed, fragmented),mainly in the central part of Mexico. Although these characteristics are factors contributing to outbreaks (Baker 1995; Dale et al. 2001; Moreno-Sanchez et al. 2012), we were not able to prove such a correlation here.

The primary concern about disturbances within forest ecosystems are those imposed by human actions that alter natural cycles. A variety of disturbances, either natural or human-induced can cause changes in the structure, function, composition, productivity and extent of forests that may contribute to the spread of pest species such as pine beetles (Sánchez-Martínez and Wagner 2009; Smith et al.2013).

Abdullah and Nakagoshi (2007) described the relation of forest fragmentation with human land-use and suggested that forest decline is a result of fragmentation caused by humans. Lu et al. (2012) found that forest fragmentation leads to forest decline as a result of the loss of carbon stocks (up to 11.9%), on top of other negative impacts(such as the loss of biodiversity). The measurement and quantification of the effects of disturbances and landscape fragmentation remain one of the most significant methodological challenges in the context of landscape analysis (Gergel 2007); however, it is essential to include the effects of disturbances, landscape fragmentation and land-use dynamics in the modelling of species distribution to provide more comprehensive, accurate results.

Conclusions

The principle of maximum entropy enabled the development of environmental or ecological niche models to predict the potential spatial distribution of D. mexicanus in Mexican temperate forest. A high value of the distribution function indicates favourable conditions for this species.The results of the present study can be used to monitor and manage D. mexicanus and serve as a model for monitoring similar species.

AcknowledgementsThe authors are very grateful for comments and suggestions from two anonymous reviewers who helped improve the original manuscript.

Compliance with ethical standards

Conflict of interestThe authors declare that they have no conflict of interest.